3D Image Processing and Visualization

On-going pioneering research and technology development in 3D image processing and visualization being carried out at the Laboratory of Intelligent and Parallel Systems (LIPS) encompasses a broad spectrum of activities. Centered around the in-house core technology of object (surface) triangulation with the ability to control both the error of approximation and the number of triangles, four categories of complementing activities, reveal the breadth and vitality of this R&D agenda.

Object Modeling: Triangular mesh representation of 3D objects is the de facto standard in 3D graphics and visualization. Accurate yet succinct representation is the primary objective of any development of this technology. Triangular mesh generation from range images, volume data, scattered surface point sets, or mathematical functions can all be carried out using cooperative agents and resource planning, achieving simultaneous optimization of quality and simplicity.

Geometric Compression: The development of efficient encoding of essential object representation based on its triangular mesh aims to achieve high compression ratio without loss of model accuracy. Dynamic schemes encoding the primitive operations in the process of model generation are also being deployed to facilitate progressive transmission across the network in the downstream applications of geometric compression.

ran.jpg (19263 bytes)

reg.jpg (18427 bytes)

Image Registration: Multiple 2D views are required to represent and visualize a single 3D object, as vital information could be occluded in any viewing perspectives. Useful models of 3D objects, thus, must match and integrate partial information from multiple views, by finding and registering the overlapping parts between views in order to construct a unique and correct 3D representation. The approach for image registration at LIPS relies on the error-controlled triangulation to discover the maximal overlap between 2D perspective views. This novel registration approach is capable of not only deriving the optimal transformation matrix that matches pair of images but also to construct a truthful 3D model from multiple 2D views of the object automatically.

Volume Visualization: Visualization of 3D spatial data and its temporal transformations is crucial to gain physical insight and to perform diagnostic analysis in scientific and bio-medical applications. A new visualization tool set is under development to allow collaborating researchers to remotely visualize and analyze spatial and temporal information embedded in a dynamic volumetric stream of captured images or computed experiments.